Adaptative segmentation of SAR images

Discusses the unsupervised segmentation of radar images. Usually the marginal distribution of each class for SAR image segmentation is supposed Gaussian or gamma, the field of classes is generally supposed stationary. the authors propose the use of different marginal distributions in order to improve the fitness of the statistic model with the data. The distributions grouped in the Pearson system provide an approximation to a wide variety of observed distributions like in radar images of the sea, ice,… To take into account the non stationarity of the class field the authors adopt a new modelization for this field. The mixture of marginal distributions and therefore the class field distribution are estimated by an adaptative algorithm. They adopt a Bayesian criterion for image segmentation. The algorithm obtained is tested on a synthetic image and also applied to the segmentation of real SEASAT scene